The principles of alliance formationbetween Konso towns

iC. R. Hallpike

/ *

Reprinted from Man vol 5 no 2 June 1970

THE PRINCIPLES OF ALLIANCE FORMATIONBETWEEN KONSO TOWNS

C. R. HALLPIKE

The Konso are a Cushitic-speaking people of south-west Ethiopia, practisinga highly sophisticated terrace and manure agriculture, their staple crop being millet.Traditionally they are organised into a number of autonomous walled towns,with an average population of about 1,500. There is very little population mobilityamong men; about 40 per cent, of women however marry men of friendly townsin the immediate vicinity. Towns are governed by elected councils of elders,but traditionally there is no authority which can impose its will on the townscollectively and, as we shall see, fighting was endemic among them. Law andorder have only prevailed since the Amhara conquest of 1897. The Konso have ageneration-grading system of the gada1 type, of which there are three differentvarieties in the northern, eastern and western regions, but these systems areessentially ritual and moral, and produce no effective cross-cutting ties to mitigatethe rivalries between the towns. They are irrelevant to our understanding ofKonso political relations. There are three regional priests who, besides officiatingat the periodic generation-grading ceremonies, act as mediators and peace-makersbetween the towns in the event of a battle. Their authority however is mysticalonly, and it is clear that their efforts have often been wholly ineffectual, and that atbest they can only provide an honourable pretext for both sides to withdraw.

Konso political alliances

Fig. i shows the relations between the towns, a continuous line denoting alliance,a broken line enmity. The map however is incomplete; I personally resided intowns 2, 9, 12 and 26, and while I was able to obtain most of the relations betweenthese towns and their friends and enemies, it was often very difficult to obtaindata on the relations between other towns where I was not residing. Thus it is notclear, for example, whether 22 and 6 are friends, enemies, or sufficiently distant tobe neutral. Nevertheless, there are sufficient data for the purposes of this article.

Only two regions are shown here; the third region, Turo, has virtually no towns,and is geographically and socially more remote from Garati and Takadi than theyare from each other. The regions are defined by their possession of a commongeneration-grading system and regional priest. But they are not political entities,and, as can be seen from the map, relations between towns are independent ofregional boundaries. Some towns are united in specially close alliances, which areringed. In such cases the alliances are ' balanced'—that is, they have all friends andall enemies in common—and all mutual relationships are as friendly as possible.I shall term such alliances 'nuclear'. This term will be discussed more fully below.There can also be other, looser, alliances which are not nuclear. Whatever thenature of the alliance, there is always one town which is dominant, being referred

Political Relations between the towns.

= town

= friend- — enemy' = approx. 1 mile

regionalboundary

zow

HI--

O

tew

ww2WOS!

o2

FIGURE I. Political relations between the towns.

260 C. R. HALLPIKE

to as apa, father, or garda, eldest brother, in relation to the other towns. The domi-nant members of alliances are also larger than their partners in almost every case.

Konso traditions on the origins of these alliances are vague, but conventionallyhold that the junior towns were founded by migrants from the senior town.In town 9 where I lived for two months there was considerable population mobilitybetween it and the other two smaller towns in the alliance. Even if there is not ahigh degree of population mobility between towns in the alliance, there is a muchhigher frequency of intermarriage between their members than with other towns.This is illustrated in the following table showing the distribution of 36 marriagesbetween the men of town 26 and the women of its neighbours, the majority ofexternal marriages being with the nuclear partner, town 25.

TABLE i.town no. of marriages

25 " 1824 723 322 2

28 2

29 I

30 I

27 I19 I

This compares with sixty marriages within town 26 itself.

As already indicated, alliances between towns are of two types: i. Unbalancedalliances of the form:

where the allies of 2, namely i and 3, are mutual enemies, potentially putting 2in the position of forsaking one alliance in the interests of maintaining the other ifI and 3 fight.

2. Balanced 'nuclear' alliances, where all the component towns share commonfriends and enemies. This distinguishes this type of alliance from others among thetowns, which are frequently of the unbalanced type. Such tight-knit alliances areringed in the map, and their frequencies are as follows:

alliances with 4 members ialliances with 3 members ialliances with 2 members 6

The significance of these figures will be considered later.Before the Amhara conquest, battles between towns were frequent. A victory

was commemorated by the erection of a stone pillar in the successful town. Sincethese are associated with named generation sets, it is possible to calculate when

ALLIANCE F O R M A T I O N BETWEEN KONSO TOWNS 26l

battles occurred to within eighteen years.2 All the stones seem to have been erectedwithin the nineteenth century, and only a few after the arrival of the Amhara.The average number of victories recorded by each town is 5-5, and this, given thenumber of towns, and the available time span, means that the battles must havebeen occurring at the rate of more than one a year throughout this period. Thefollowing table lists the victories of town 12:

TABLE 2.

town datebetween 12 and 18, 19 1827-45

13 1845-6313 1845-6318, 19 1863-816 1863-8114 1881-9913 1899-191713 1935-53

According to accounts given by elderly informants, battles were not usuallypushed to extremes, that is they did not normally result in the destruction of thedefeated town. Customarily, the warriors advanced to meet each other across thefields, meeting in the open, and retiring when one side was worsted in the exchangeof spears and stones, though in the fiercest battles towns were sometimes burned andlooted. Frequently the occasions of the conflicts were trivial. Thus I was toldthat in about 1875 a dance was held at which a fight broke out between the men of25 and Ibale, a town no longer existing, but which was near town 27. The next daythe men of Ibale attacked 25 and partially destroyed it, whereupon the remnantsof 25 and their ally 26 attacked Ibale, completely destroying it. As another example,26 and 28 fought because 28 stoned the goats of 26, which were straying onto thefields of 28 and eating the crops. Again, Ibale, before it was destroyed, denied theuse of a path to the lowlands to 26, who successfully asserted their rights by battle.Battles therefore seem mainly to have resulted from pinpricks and minor griev-ances, and not from competition for the necessities of life. A particularly revealingcase of this type occurred a few years before I came to Konso. As can be seen on themap, 12 is friendly with 9, 10 and n; and 7 and 8; but towns 9, 10 and n on theone hand and 5 and 6 on the other are enemies of 7 and 8. The wells of 7 and 8 randry so they asked 12 for permission to use theirs. Town 12 agreed, but 9, 10 and nobjected violently, and accused 12 of betraying them, though they themselves didnot need to use the latter's wells. A pitched battle between 9, 10 and n, and 7 and8 was only prevented by the police and the intercession of a Konso teacher at themission. The result was that 9, 10 and n, already being enemies of 7 and 8, brokewith 12; while 5 and 6 sided with 9, 10 and n against 12.3

We thus have a situation in which towns are prepared to fight, alone or inalliance with other towns, over relative trifles; there is cultural homogeneity, suchthat all parties fight and ally within a common body of accepted norms; battlesare not normally followed by the complete destruction of opponents, and are notfought to control vital economic or ecological resources, such as trade routes orwells, nor for political power. While wells, paths to grazing land, or fields, maybecome involved in disputes, in the sense of being instruments of provocation,

262 C. R. HALLPIKE

I must emphasise that they are not, in the usual way, the objects of competition,such that their control is the prize of victory. The only exception to this generalisa-tion which I know of is the frequent conflict between 12 and 13 over a plot ofland midway between them. The readiness of the Konso to fight reflects theirextremely high evaluation of military prowess and virility, and not economicscarcity.

While, as I have said, there is a ritual and moral unity within the regions, thepattern of alliance and enmity has no relation to the grading systems, the regionalpriests or the sacred drums, which are ritually associated with peacemaking.Nevertheless, while town relationships are apparently anarchical in operation,in relation to the norms and comprehension of the Konso themselves, I hope todemonstrate that the formation of alliances, and in particular of nuclear alliances,is actually governed by a few fundamental principles.

The explanatory model

The object of this article is to show that explanations of the nuclear alliances interms of migrations is unnecessary, and implausible, since migration is usually asign of friction rather than of friendship, and that a limited number of assumptions,of wide generality, combined with random processes, will achieve the same results.

D5

FIGURE 2. The model network.

ALLIANCE F O R M A T I O N BETWEEN K O N S O TOWNS 263

I therefore constructed a model network which would allow towns to havevarying numbers of allies, for relations to cease with distance, and for allies to bebrought into the conflicts. On this basis, rules were formalised governing theformation of alliances, and a procedure established whereby a sequence of battles,selected by random processes, could be simulated, until the relations between thetowns permitted no further change. I refer to each sequence of battles ending insuch an equilibrium situation as a 'game'. I refer to the total rules system as 'themodel'. The model incorporates two fundamental assumptions for which I haveno substantive evidence from field observation, yet which are inevitable forany decision-making procedure of the kind we are considering, when towns havein many cases to choose between such of their allies as are fighting. My assumptionsare i) that every town ranks every other town with which it has relations accordingto some scale of preferences, and 2) that these preference ratings can be assym-metrical e.g.

A1 2B

Here A ranks B higher than B ranks A. These assumptions are of extreme sim-plicity, but they have far-reaching consequences.

The network which was established, as in fig. 2, contains fourteen towns,A-N, all joined in a fixed set of relations. It will be observed that the number ofrelationships varies from town to town, A thus having four relationships, B six,and so on. The degree of enmity/friendship between any two towns is expressed inscale 1-5, as follows:

1 . . . . very friendly2 . . . . fairly friendly3 . . . . neutral4 . . . . fairly unfriendly5 . . . . very unfriendly

An odd number of rankings is necessary to express neutrality; only three ranksis too crude, while seven or more is implausibly complex, at least in politicalrelations.

In playing the game the evaluations according to this scale are then distributedamong the towns at random by use of a random number table. Initially, however,in order to give as little bias as possible to the network in any direction, only 2, 3,and 4 evaluations are allotted, and no z's or 5's, to keep the pattern of evaluationsas close as possible to neutrality while retaining any possibility of choice. Fig. 3will make this clearer.

In fig. 3 which represents a possible initial allocation of evaluations at thebeginning of a game, A rates B as 2, C as 3, G as 4, E as 3. In turn, B ranks A as 3,C ranks A as 3, G ranks A as 4, and E ranks A as 4, and so on. These relationshipsare then written down in a list and numbered, every time the game is played, asfollows:

1. A=B22. A=C33- A=G44- A=E3and so on.

264 C. R. HALLPIKE

FIGURE 3. The starting position.

The exclusion of I's and 5'$ in the initial allocation of ratings also means that onecannot have extremes of assymmetry in mutual ranking e.g.

A1- -5B

Towns therefore can only differ by two points in their mutual evaluations in thismodel.

Having listed all the binary relationships (in this particular network, 94) aparticular relationship is chosen from the list by use of a random number tableagain. Let us suppose that it is relationship 2, A, C. A battle is supposed to occurbetween the members of the relationship selected in this way, the only exception,as a concession to the realities of life, being that no battle can occur between townswhich mutually rank each other i-i. This limitation however cannot apply atthe beginning of a game, since no I's are allotted. There are thus two principalopponents—in this example, A, C. But since other towns have relations with A andwith C they will also be drawn into the conflict. The consequent shifts in alliancesare governed by the following rules:

i. The potential ally or opponent of one combatant must also have a directrelationship with the other combatant. Thus, in the battle between A and C, the

A L L I A N C E F O R M A T I O N BETWEEN KONSO TOWNS 265

potential allies of both are B, E and G. I and F have a relationship with C, but notwith A directly. In real life F would have to decide how it rated A as well as C;and in the model network here it is assumed, as might well be the case in real life,that A is too distant for F to have relations with it.

2. Potential allies or enemies of combatants choose between them as follows:a. by their own rating of the combatants. For example, B rates A as 3, C as 4,

so B becomes an ally of A in the fight. E however rates A as 4, and C as 2, and sobecomes an ally of C. But if potential allies of combatants rate the combatantsequally;

b. the choice is made on the basis of the combatants' rating of the potential ally.For example, if C and G were fighting, E would join G rather than C, because Grates E as 2, while C only rates it as 3. But if the ratings are again equal;

c. if, for example, C and G both rated E as 2, the potential ally remains neutral.This is particularly important in the special case where three towns are related asfollows:

A will remain neutral in this case.3. Once battle is joined by the principal combatants and their allies, the ratings

between the principal combatants themselves, the combatants and their allies, andbetween the allies, will all change, according to the following rules:

a. The principal combatants decrease their mutual ratings by one, in thedirection of enmity,

b. allies on the same side increase their mutual ratings by one, in the direction offriendship;

c. allies on opposite sides decrease their mutual ratings by one. It should benoted that, while towns which rate each other i-i cannot fight directly, they maybe allies on opposite sides, e.g.:

Here, A supports D, B supports C; thus by rule 3 (iii) above, A's and B's mutualratings drop to 2-2.

Thus it should now be clear why, in the conflict between A and C in fig. 3,the final rankings of all interested parties will be as in fig. 4. Here A and B are muchmore closely linked than before the battle, and E, C and G are also much closer.

It is of course possible that the original allies might become the chief disputants

266 C. R. H A L L P I K E

FIGURE 4. Final ranking from fig. 3.

ill a battle, but there is no reason to think that this would make any difference tothe resulting rankings after the battle, and it is therefore irrelevant for the model.

Nor is there any reason to suppose that victory or defeat would be a relevantfactor in the formation of the final ranking. If, say, A and E are defeated by C andG, neither A nor E has any reason to rate the other lower than if they had been thevictors. This is related to the basic nature of the political relationships involved.For example, if we were simulating intrigue at a royal court, and A and E weredefeated by C and G in competition for some sinecure, both A and E might beattracted to C and G, as new centres of power, and repelled from each other, aslosers, but Konso political conflicts are not competitions for power. They simplyreflect the existing patterns of alliance and enmity, and are occasioned by relativelytrivial disputes. As we saw, battles do not place any resources on which militarypower depends at the disposal of the winner. Konso towns are thus in a quitedifferent situation from that where each victory increases the possibility of futuredefeat.

Again, if it were customary for victorious towns to take over the assets ofconquered towns, this might lead to subsequent quarrels among the victors overthe spoils, a factor which would have to be allowed for in the rules, but which does

A L L I A N C E F O R M A T I O N BETWEEN KONSO TOWNS 267

not in fact seem to occur. The outcome of a battle is not therefore relevant to themodel.

Finally, the end of a game occurs when all relationships reach a stage of beingeither i or 5, and consequently when no further changes in alliances are possible.In practice one finds that one or two relationships sometimes remain in the 2-4range, but these are always isolated cases which cannot affect any other relation-ships in the network. (I deal with some of the undecided cases below.)

These then are the rules of the game, and the reader may concede that while itinevitably has a certain ' stiffness' in comparison with real life, its basic assumptionsare both realistic and very simple, with a minimum of gratuitious suppositions.

Given these rules, and the arrangement of the network, the nature of the out-come is not apparent to observation or common sense. A number of conclusionsmight seem possible—a state of undifferentiated enmity, or friendship, or totalconfusion, with all relationships unbalanced, the latter being perhaps the mostlikely. However the game was played three times, using different sets of randomnumbers on each occasion; in each case the results were of the same type—a smallnumber of nuclear alliances with all friends and enemies in common, bearing in thisand other respects a striking resemblance to the actual situation in Konso. I will

A — 2 3— B

M —2

FIGURE 5. Game i: initial allocation of rankings4—M.

268 C. R. HALLPIKE

A —1

M —1 1— N

FIGURE 6. Game I: final distributions of rankings.

discuss the results of the first game in detail, and then give the results of the othertwo.

The distribution of ratings at the beginning of game i is given in fig. 5, and thedistribution of ratings at the conclusion in fig. 6. It will be seen that in all but threecases rankings are i or 5. I and M rank each other as 3, but since they have allenemies in common, while M's principal ally, N, cannot come into direct allianceor conflict with I, the future of their relationship cannot be predicted strictly interms of the rules of the game. However, if I and G fight one another, or M and G;or I and K; or M and K; I and M will be allies, and their mutual rating will risetowards i. Only if a direct conflict occurs between I and M will their mutualrating fall. In terms of probability therefore there is an 80 per cent, chance of theirending as i-i allies. The other two cases, N=L4, L=F4, can only go to N=L5,L=F5, given sufficient time. The reasons for this are as follows. In any furthercombat involving N and L or their allies, they will take opposite sides. If F andJ or F and K fight, L supports J or K, so in either case L's rating of F goes from 4to 5, as it will if F and L fight.

In another case, depicted in fig. 7, taken from the third game, there is a morefluctuating relationship. Here, C=l4, I=C5. If C and G fight one another, orI and G; I and C are allies, and their mutual ratings improve to C=l3, I=C4. If

A L L I A N C E F O R M A T I O N BETWEEN KONSO TOWNS 269

D

NFIGURE 7. One configuration from game 3.

however I and J fight, I rates C as 5, and C's rating of I also drops to 5. Thissituation is irreversible, since if I subsequently fights G, C remains neutral, becauseit now rates I as 5, not 4 as previously. Thus here the developments of the relation-ship between I and C depend on the battles, and their order selected by the ran-domising process. It is possible therefore to have relationships which fluctuatebetween friendship and enmity, and which within the limits of the game are leftunresolved, though in principle, if the game were continued long enough, theoutcome even here is predictable. These undecided situations however are veryfew, and not a basic anomaly.

In each game therefore, with these rare exceptions, the end is indicated by acondition in which all towns rank each other as i or 5, after which no changes arepossible by the rules of the game, since no battles can take place between townsrating each other as i-i, while in such cases as:

2JO C. E. HALLPIKE

FIGURE 8. Game i: pattern of alliances.

if B and C fight, A remains neutral. But B and C already rate each other 5-5, sowhile battles can still occur, they will not change the pattern of alliances.

Fig. 8 shows the pattern of alliances more clearly than fig. 6. The towns arelinked in a complex chain, in which only one nuclear pair, M and N is visible, aswell as an isolate, I. M and N are defined as a nuclear pair because a) they rankeach other as i-i, and b) because they have all enemies in common, and thereforetheir alliance contains no contradictions. On this basis, if we examine fig. 8 moreclosely, we can find further sets of nuclear relationships, as shown in fig. 9.

Towns linked in nuclear relationships must therefore have the followingcharacteristics:

1. They must all be mutually allied, as, for example, E, C, G; in fig, 9. A, B, F;however do not form a nucleus of 3, since the form of the network does not permita direct relationship between A and F. This marks a slight difference between themodel and reality, since in actual life we might have a relationship in this casebetween A and F, but according to the rules of this game we must describe A, B;and B, F; as two separate nuclei of two.

2. If any town not in the nucleus has relations with more than one memberof the nucleus, these relations must all be consistent, that is, all hostile or all

A L L I A N C E F O R M A T I O N BETWEEN K O N S O TOWNS 271

D

H

FIGURE 9. Game I: nuclear alliances.

friendly. But such a town need not have relations with all the towns of the nucleus;e.g. in fig. 8 only C and G have relations with J, not E, but this does not preventC, E and G forming a nuclear alliance. (Here again, the rigidity of the model forcesa slight departure from the conditions of real life.) As we have seen, this means thatthere are no contradictions within the alliance. For example, in fig. 8 we see thatC, D and G all hate B and F, but C and G like J, while D does not, and this wouldintroduce a contradiction into the relationship between C and D, and preventthem forming a nuclear relationship. In the same way, C and J are both allies, butJ could not join the alliance of C and G because C is the ally of D, which is J'senemy. It might be asked however if E, G and K could not also form a nuclearalliance fulfilling all the above conditions. The answer is, yes, but this would leaveout the relationship of C to E and G, and J to K. We have therefore,

3. to construct nuclear alliances using as many potential partners as possible.This may seem an arbitrary rule, but in fact, since there are necessarily more ways ofrelating only some pairs of nuclear allies than there are of relating all potentialnuclear allies, this rule ensures that the nuclear relationships finally selected are aslittle the result of arbitrary choice as possible.

Having explained the principles by which the nuclear alliances are established,

272 C. R. HALLPIKE

3— B

4— N

FIGURE 10. Game 2: initial allocation of rankings.

therefore, I now show the beginnings and ends of the second and third games infigs. 10-17.

The interpretation of the model

The reasons for the formation of the nuclear alliances are as follows:I. whether towns are allies in the final outcome depends not only on their

mutual ratings to begin with, but on the total nexus of their relations with theirneighbours, in which structural balance is crucial, and on the manner in whichthese are modified by the random order of battles. (It is important to bear in mindthat the order of the battles is crucial to the final result.) This inter-dependence oftowns is the basis of the cumulative process, whereby;2. the existence of unbalanced relationships forces towns in most cases to takesides in a combat against another ally, early in a game, and;3. the more two towns fight on opposite sides, the greater their tendency to fightin the future, and, the more often they fight on the same side, the stronger theiralliance becomes. This tendency to alliance or enmity becomes fixed when;4. a 'point of balance' is reached in any nexus of relations. After this point eachnew battle inevitably drives the towns in question towards greater and greaterenmity, or closer and closer alliance.

A L L I A N C E F O R M A T I O N BETWEEN KONSO TOWNS 273

A —1

FIGURE n. Game 2: final distributions of rankings.

Thus, at the end of the game, towns which are closely contiguous are likelyeither to be i-i, or 5-5, because they will have been involved most often in situa-tions of conflict and decision, such that relationships in their immediate neighbour--hood will tend to have been resolved one way or another. Only in the case of moredistant relationships may ambiguities persist, as between C, D and J in fig. 8.D has no direct relations with E or G (C's nuclear allies) or with K, J's ally; andthus the ambiguities have never been resolved.3 This total process permits longchains of alliances, some of whose members are enemies of other members. Weobserve the same phenomenon in the map of Konso political relations (fig. i).

The appearance of nuclear alliances, and chains of allies which embody con-tradictions, are not the only resemblances between the model and Konso reality.

In the three games played we find that the majority of alliances have only twomembers, and that the distribution of alliances between 2, 3 and 4 members is asfollows:

No. of members2

34

game i

32

O

game 2

3io

game 3

3oo

total

93o

274 C. E. H A L L P I K E

FIGURE 12. Game 2: pattern of alliances.

In the map of Konso towns we see that the distribution of alliances between 2, 3and 4 members is as follows:

No. of members2

34

Konso6iI

A direct numerical comparison between the Konso situation and the model isimpossible, because of the greater number of Konso towns, and the differentpattern of their distribution. But it is possible to show that the high number ofalliances with two members in the model is not in accordance with the opport-unity of forming them. In the model the total number of binary links is forty-seven, of triangular links fifty-six, and of quadrangular links (where all membersare linked, as are GIJK, for example) fifteen. On the basis of probability thereforewe would expect alliances of 2 in 40 per cent, of cases, of 3 in 47 per cent, of cases,and of 4 in 13 per cent, of cases. The actual distributions, in table 3 below, show thatthe model is very noticeably closer to the Konso distribution than considerationsof probability would lead us to expect, though, of course, the figures involved arevery small and statistically inadequate.

A L L I A N C E F O R M A T I O N BETWEEN K O N S O TOWNS 275

H

M NFIGURE 13. Game 2: nuclear alliances.

TABLE 3.

no. in alliance by probability in realityin model in model

2 40% 75%3 47% 25%4 13% o%

in Konso

75%12-5%12-5%

There is a good reason however why alliances of two should be commonerthan alliances of three or four. This is that alliances of two have fewer relationswith other towns than alliances of three or four. Hence it is much easier for themto avoid contradictions in their alliances than is the case with alliances containingthree or four members.

Another noticeable resemblance between the model and the Konso situation isthe percentage of towns in nuclear alliance in both cases. In the map, if we ignoretowns 14, 15, 16,17, 35 and 36, about whose relations I know very little, this leavesthirty towns, nineteen of which, or 63 per cent., are in nuclear alliance. In thethree games played, the following proportion of towns in nuclear alliances canbe observed: 11/14, 8/14, and 6/14, or on average 60 per cent. Such numericalcorrelations are not, of course, to be taken too seriously, but at least they indicate

276 C. R. H A L L P I K E

A —2

FIGURE 14. Game 3: initial allocations of rankings.

that the results of the model and the Konso situation are reasonably similar in thisrespect also.

Finally, we have seen that in two games isolated towns with no friends occurred,like town 31 in Konso, whose alliance with town 12 has no military significance.

The results of the model thus resemble the realities of the Konso situation in thefollowing ways:

1. There are similar proportions of nuclear alliances.2. The proportions of 2, 3 and 4 member alliances are also similar.3. There are many unbalanced relationships.4. There are long chains of alliances which are not nuclear.5. Isolated towns can occur.

In one important respect, however, the model has no relation to the Konsosituation, since it takes no account of the disparity in size between leaders and othermembers of the alliances. Indeed it might be suggested that nuclear alliances couldbe explained more simply by the attraction which large towns exert upon theirsmaller neighbours. In the first place, however, such an explanation does not fit thefacts. For if this attraction of larger for smaller were the sole factor in the formation

ALLIANCE FORMATION BETWEEN K O N S O TOWNS 277

FIGURE 15. Game 3: final distributions of rankings.

of alliances, how is it, for example, that 9, the largest town in Konso, should nothave drawn 7 and 8 into its orbit, since they are no further than 10 and n? Again,12 is considerably larger than. 13, yet they are bitter enemies, and 13 has formed analliance with 9,10 and 11 which are considerably further away than 12. Also, beforeits partial destruction, 25 was apparently a very large and powerful town, and thesestatements are supported by the visible remains and the large number of stones ofvictory. Yet towns 25 and 26 are closer allies than 25 and 24, though 24 is as nearto 25 as 26 is. More generally, the 'large town magnetism' theory is conspicuouslyunable to explain why there should be a preponderance of two-member alliances;if the theory were correct we should expect a majority of four or five memberalliances. The theory also assumes that the ratios of town sizes were the same in thepast as they are today. The most we can say about the status of large towns innuclear alliances is that, other things being equal, large size will be a factor likelyto have a favourable influence on the ranking of potential allies, though, on theother hand, it may also be a cause of fear, and hence repulsion.

There remains a final possible explanation—migration, which is the one givenby the Konso themselves. There is very little evidence in most cases relating to theorigins of the towns, but 12, for example, was originally formed from five smaller

278 C. E. H A L L P I K E

FIGURE 16. Game 3: pattern of alliances.

settlements in the vicinity, while 26 is supposed to have been founded by menfrom 31 as well as from 25, although 31 and 26 are long-standing enemies. Ingeneral, family histories suggest that there has been considerable mobility ofindividual families, but the evidence of the present ground-plans of town wallsmakes it clear that towns normally do not undergo a process of fission when theyexpand, but simply grow larger. Moreover, as I remarked earlier, if sufficientpeople to found a separate town were to split off from the parent body, it is likelythat this would only have been as the result of a serious quarrel. It is very possiblethat migrations and stochastic processes are responsible for the form of the internalorganisation of the towns, but this must be left for a separate analysis. In general,both the 'large town magnetism' and the migration theories demand many morehistorical assumptions than the model requires, and such conclusions as one candraw from them do not fit the facts of the Konso situation so well. These theoriesalso lack the systemic characteristics of the model. This is not to deny however thatthey may have had some influence; I am simply saying that by themselves they areinsufficient, and demand a number of improvable historical assumptions.

The value of simulations of this sort (which are ideally suited to the computer,though in this case the model was too small to justify one) is that they rely on

ALLIANCE FORMATION BETWEEN KONSO TOWNS

B

279

FIGURE 17. Game 3: nuclear alliances.

assumptions of a very simple nature, which are readily generalisable to othersocieties, since problems of conflict, alliance, choice and ranking are fundamentalboth at the societal and personal levels. For example, in experimental psychologymuch work has been done on the problems of structural balance (Cartwright &Harary 1956), that is, on how contradictions in networks of personal evaluationsare resolved, and these problems have important relations to the branch of mathe-matics known as 'graph theory' (Flament 1963). There is reason to hope thereforethat some aspects of the model presented here may be useful in explaining insti-tutional phenomena in other societies besides that of the Konso.

NOTESI am obliged to Dr David Elliott for introducing me to the literature on structural balance and

graph theory, and for a number of valuable discussions on themes raised in this article. I alsoreceived some very useful suggestions from Dr James Fox and other members of the Depart-ment of Social Relations at Harvard University, where this paper was presented at a seminar.For a full account of the Konso see Hallpike (in press).

1 That is, that found amongst the Galla and other Ethiopian peoples, where one's positionin the system depends not on chronological age, but on the position of one's father in thesystem.

2 The generation-grading system of Garati region, in which town 12 is situated, has a cycleperiod of eighteen years. Since the historical sequence of the generation-sets is known byinformants it is possible to obtain the approximate dates of battles in this way.

28O C. R. HALLPIKE3 The map shows the relations between 9, 10, II and 12; and 5 and 6; before this quarrel.4 In this model, it sometimes happens that some towns, especially M and N, occur less

frequently in the random number selection than others, and therefore their relationships donot always have time to be resolved by the end of the game.

REFERENCES

Cartwright, D. & F. Harary 1956. Structural balance: a generalization of Heider's theory.Psychol. Rev. 63, 277-93.

Flament, Claude 1963. Applications of graph theory to group structure. Englewood Cliffs, N.J.:Prentice-Hall.

Hallpike, C. R. in press. The Konso of Ethiopia: a study of the values of a Cushitic people. Oxford:Clarendon Press.

Sweetser, D. A. 1967. Path consistency in directed graphs and social structure. Atn.J. Social. 73287-93.